Mortality at 30 and 90 days, stratified by age.
Deaths in the first year after colorectal surgery.
Visser BC, Keegan H, Martin M, Wren SM. Death After ColectomyIt's Later Than We Think. Arch Surg. 2009;144(11):1021–1027. doi:10.1001/archsurg.2009.197
Clinical outcomes are increasingly subject to objective assessment and professional accountability. Informed consent relies on accurate estimation of operative risk. Current scoring systems for assessment of operative mortality after colorectal surgery (CRS) almost uniformly report 30-day mortality and may not represent true risk.
University-affiliated Veterans Affairs Medical Center.
All patients who underwent resections of the colon and/or rectum (as the principal operation) at a single hospital whose data are captured in the Veterans Affairs National Surgical Quality Improvement Program (VA-NSQIP) database from January 1, 2000, through December 31, 2006.
Main Outcome Measures
Mortality at 30 days and 90 days.
The VA-NSQIP cohort included 186 patients who underwent CRS, including 148 patients who underwent elective procedures (79.6%) and 38 patients who underwent emergency procedures (20.4%). All but 8 patients were men, with a median age of 67 years (range, 26-92 years). Laparoscopic operations comprised 24.2% and open operations comprised 75.8%. Most (60.8%) were performed for neoplasms. The actual 30-day mortality rates (all, elective, and emergency procedures) were 4.3%, 1.4%, and 15.8%, respectively. These rates closely mirrored the calculated VA-NSQIP risk-adjusted observed-to-expected ratio for 30-day mortality (4.8%, 1.8%, and 18.2%, respectively). However, mortality at 90 days increased substantially to 9.1%, 4.1%, and 28.9%, respectively.
The 30-day mortality significantly underreports the true risk of death after CRS. The 90-day mortality rate should be included as a standard outcome measure after CRS because it serves as a better estimation of risk for counseling patients.
Among the central and immutable elements of medical ethics is the principle that physicians must obtain informed consent from their patients before initiation of any intervention.1 This is particularly true in surgery, where information about the risks and benefits of a procedure is relevant to the decisions of patients as to whether to proceed with the operation and where and from whom they will seek care.2 Simultaneously, there is now, more than ever, a growing emphasis in medicine as a whole on outcomes, quality, and patient safety. Again, this is perhaps more true in surgery than any other arena, where the interventions are the most invasive and the results, both positive and negative, most notable.3,4
Certainly no measure of quality is more concrete, easily quantifiable, or important to patients than postoperative mortality. Accurately quantified and published mortality rates can (1) permit comparisons among surgeons and institutions, (2) serve as a catalyst for quality improvement,5 and (3) facilitate accurate counseling of patients with regard to the risks of any operation.6 Yet there remains no single definition of the duration of time after surgery that encompasses what we call a postoperative death. Although various authors who have described the spectrum of operations have arbitrarily truncated postoperative mortality during the inpatient stay (in-hospital) or at 30, 60, or 90 days, certainly the most common definition remains 30-day mortality. This marker serves as the index in the bulk of articles that describe outcomes after colon resection,7- 9 such as the Veterans Affairs National Surgical Quality Improvement Program (VA-NSQIP)10 and the more recent American College of Surgeons Patient Safety in Surgery Study.11 However, there is growing recognition that mortality associated with surgery extends beyond 30 days for a variety of operations.12- 14 Although 30-day mortality certainly allows comparisons among institutions in efforts to improve outcomes, it may not serve as an accurate number to quote to patients during discussion of the risks of an operation. We hypothesized that the 30-day mortality rate significantly underreports death after colorectal surgery.
This study included all patients who underwent colorectal surgery (CRS) at our hospital whose data were captured in the VA-NSQIP database from January 1, 2000, through December 31, 2006. The VA-NSQIP prospectively gathers extensive preoperative, intraoperative, and postoperative data on a random sample of general surgery operations, with approximately 50% catchment in our center. The details of the VA-NSQIP have been described in detail elsewhere.15 Open and laparoscopic colorectal resections were identified based on the Current Procedural Terminology code, with the inclusion of any partial colon and/or rectal resection (open: 44140-44141, 44143-44146, 44160, 45111-2; laparoscopic: 44204-44208), abdominoperineal resection (open: 45110; laparoscopic: 45395), and total colectomy (open: 44150-44158; laparoscopic: 44210-44212). Patients were excluded if the colon resection was performed in the context of another major operation (eg, left colectomy performed with en bloc resection of the tumor in the tail of the pancreas). Among the patients included in the study, deaths were identified via the VA-NSQIP itself (if they occurred within 30 days) and the Veterans Affairs (VA) Computerized Patient Record System (CPRS). The CPRS is routinely cross-referenced to the Social Security death registry, so deaths are accurately captured for this patient population. The mean (SD) length of follow-up for the entire cohort was 2.81 (1.7) years. Deaths were categorized as within 30 days, within 90 days, or in-hospital (before discharge after index operation). The events preceding death, as well as the causes of death, were determined by review of notes within the CPRS and death certificates.
Continuous data were expressed as mean (SD). Groups (eg, emergency procedures vs elective procedures) were compared by means of the Fisher exact test for nominal variables and the t test for parametric variables. The VA-NSQIP uses stepwise logistic regression modeling to predict the probability of complications and death (within 30 days) for patients who undergo general surgery. This is expressed as an observed-to-expected (O/E) ratio. The exact model specifications are not available publicly, although a simplified version is as follows:
where f(x) = −7.89 − (0.62 × albumin) + (0.65 × American Society of Anesthesiologists [ASA] class) + (0.01 × blood urea nitrogen) + (1.03 × disseminated cancer) + (1.01 × ascites) + (0.03 × age) + (0.56 × emergency).
Therefore, an O/E ratio was available for each patient and can be calculated for the cohort as a whole. These were compared with the observed mortality. P<.05 was considered statistically significant. This study was approved by the Stanford University institutional review board.
The VA-NSQIP cohort included 186 patients who underwent CRS, with the inclusion of 148 who underwent elective procedures (79.6%) and 38 who underwent emergency procedures (20.4%). All but 8 patients (4.3%) were men, with a median age of 67 years (range, 26-92 years). The indications for surgery are given in Table 1 and are typical of the VA population. The most common indication for surgery was cancer of the colon (n = 79; 42.5%) or rectum (n = 21.2; 11%). The percentage of operations performed laparoscopically was 24.2%, whereas open procedures comprised 75.8%.
The observed 30-day mortality rate for the entire cohort was 4.3% (n = 8), which breaks down to just 1.4% among patients who underwent elective procedures (2 of 148) but 15.8% for patients who underwent emergency procedures (6 of 38). These numbers closely mirror the expected 30-day mortality rate given by VA-NSQIP (which uses preoperative variables for its risk-adjusted modeling) for the 3 groups (4.8% for the entire group, 1.8% for patients who underwent elective procedures, and 18.2% for those who underwent emergency procedures). Seventy patients (37.6%) experienced at least 1 perioperative occurrence as defined by the VA-NSQIP criteria, such as superficial surgical site infections, pulmonary emboli, and unexpected intubation.
Notably, mortality at 90 days increased markedly. The mortality rate for the entire cohort doubled to 9.1% (n = 17), with the inclusion of 4.1% for those who underwent elective procedures (6 of 148) and 28.9% for those who underwent emergency procedures (11 of 38). Patients who died within 30 days were similar to those who died in the subsequent 60 days (Table 2), including with respect to comorbidities and preoperative laboratory studies (eg, blood urea nitrogen, albumin). We included in Table 2 (and subsequent tables) the mean VA-NSQIP probability of complications and probability of death for each group of patients. Although these are calculated values, the model includes a number of preoperative variables that have been previously demonstrated to predict morbidity and mortality. Thus, they serve as useful shorthand “yardsticks” to compare groups of patients. Interestingly, although the major individual predictors that go into the VA-NSQIP model (eg, age, albumin level) are not different between the patients who died at less than 30 days and those who died between days 30 and 90, the VA-NSQIP probability of complications and death suggests that those who died within 30 days were “sicker” going into the operation, with P values that are very close to the <.05 threshold.
Patients who died within 90 days of surgery are compared with the remaining patients in Table 3. Those patients who died within 90 days were significantly older, had a worse preoperative metabolic profile (eg, albumin level, renal function), and a higher mean ASA score. However, those who died were more likely to have undergone emergency surgery (and thus less likely to have undergone surgery for cancer), and the operative time for these patients was somewhat shorter. We also investigated in-hospital mortality (death during the index admission) as an alternative to 90-day mortality. The overall in-hospital mortality rate was 7.5%. Four patients died within 90 days but after discharge (1 immediately after a routine clinic visit and 3 in hospice care after difficult postoperative courses). One patient died at day 120 after a long and difficult hospital course.
Mortality increased with age (Figure 1). The mortality rates for patients younger than 60 years (n = 66) was 0% and 3% at 30 and 90 days, respectively, compared with 5% and 8% among patients aged 60 to 69 years (n = 40), 4% and 6% for those 70 to 79 years (n = 47), and 12% and 27% for patients 80 years and older (n = 33) (P < .05, comparing age ≥80 years to each younger group).
The indication for surgery, operation, and postoperative course of those patients who died within 90 days of their operation are summarized in Table 4. Typically, these were elderly (≥65 years of age), high-risk patients who had a series of discrete complications during a prolonged intensive care unit course, which ultimately culminated in multiorgan failure and death. Notably, those patients who died between the 30-day and 90-day marks are essentially indistinguishable from those who died before the 30-day mark quantitatively (Table 2) and qualitatively, with respect to the nature of their postoperative course (Table 4). All but 2 of the patients who died before 90 days of postoperative complications. One patient died on day 25, just 2 days after a routine clinic visit during which he had been described as “recovering well.” His family declined an autopsy, although his physicians thought the death was likely owing to a pulmonary embolus. Another patient, who underwent palliative exploratory laparotomy and bowel resection for an obstructing colon cancer with carcinomatosis, died of progression of his cancer on day 86.
Emergency operations were associated with the highest postoperative morbidity and mortality rates (Table 5). The patients who required emergency surgery were, not surprisingly, “sicker.” They had a worse metabolic profile (eg, albumin level, renal function) and a higher mean (SD) ASA score compared with patients who underwent elective surgery (3.4 [0.7] vs 2.7 [0.6]). After the operation, the emergency patients stayed longer in the hospital, trended toward more complications, and had a mortality rate that was 7 times that of patients who had undergone elective procedures.
In total, 28 patients died during the first year after their CRS procedure (Figure 2). All but 1 of the deaths attributable to the operation itself occurred within 90 days. The deaths from progression of cancer within 6 months occurred in patients who had undergone palliative operations in the setting of known stage IV disease or patients who were found to have carcinomatosis at the time of their operation.
Since the publication of the scathing 1999 report from the Institute of Medicine, To Err Is Human: Building a Safer Health System,16 there has been ever-growing emphasis on objective evaluation of surgical outcomes as a means to continuous quality improvement. Certainly there is no more basic outcome after surgery than mortality—indisputable, quantifiable, and greatly important. Thirty-day mortality is certainly the most common definition of postoperative mortality in the surgical literature, likely principally because of the ease of following up patients for this short duration. However, in various articles that describe different operations, mortality has also been reported as in-hospital (with the inclusion of deaths only during index admission), 30-day-plus-in-hospital,17 60-day,18 90-day,13 120-day,19 or even 180-day.20 Nonetheless, the 30-day mortality rate for any given operation is the number most often used by physicians, whether to compare outcomes among various institutions or to quote a number to a patient during preoperative counseling.
Fundamentally, little is surprising about the 30-day mortality rate after CRS in our cohort. The observed mortality rate for all patients was 4.3%, with low mortality rates for those who underwent elective procedures (1.4%) compared with those who underwent emergency operations (15.8%). These numbers are consistent with those found in the literature broadly (1.5%-7% for elective colectomy and 12%-25% for emergency colectomy7- 9,21,22) and specifically within the VA system itself.23,24 The risk factors for mortality match those demonstrated in previous studies—advanced age, low albumin level, poor renal function, high ASA score, and emergency surgery. The observed mortality in our small cohort also closely mirrored the risk-adjusted mortality predicted by the VA-NSQIP. The VA-NSQIP database, by virtue of the detail and quality of the data and sheer number of cases, is the most powerful tool available to date for the measurement of surgical outcomes.
Notably, however, the mortality rate for our cohort at 90 days approximately doubled: 9.1% overall, with the inclusion of 4.1% for patients who underwent elective operations and 28.9% for those who underwent emergency operations. The patients who died between days 30 and 90 are relatively similar to those who died in the first month, although perhaps the former were overall not quite as “sick” going into the operation (Table 2). Table 4 is included to more vividly illustrate the characteristics of the patients themselves. Most were elderly or high-risk candidates, and most of the operations were performed for life-threatening emergencies. Most surgeons who perform colorectal resections have undoubtedly had similar patients and are likely not surprised by the outcome of these operations. Thus, it is not the “who” or “why” concerning these deaths that is surprising but perhaps the “when.” As depicted in Figure 2, the number of deaths after CRS that are attributable to the operation itself continue to increase after the traditional 30-day mark. It appears to be elderly patients who are most vulnerable to the slow decrease in health during a protracted hospitalization that leads to these late deaths. Among patients 80 years or older, the mortality rate increased from 12% at 30 days to 27% at 90 days (Figure 1), and 6 of the 9 deaths that occurred between days 30 and 90 happened in patients 79 years or older.
We hoped, in analysis of the deaths in our cohort, to identify a subset of patients who might be identified preoperatively to be better served without surgery. However, when we stratified patients who are intuitively high risk by the predictors that are easily available to physicians (eg, octogenarians who require emergency operations), we did not find a distinct group whom we would a priori exclude from consideration for surgery. Interestingly, however, 5 patients had a VA-NSQIP–estimated probability of death of 0.5 or 50.0% or more. All were in desperate situations: 3 had acute colonic ischemia, 1 had a perforated colon owing to a large bladder tumor, and 1 had fulminant Clostridium difficile colitis. Four of 5 died within 90 days and the fifth died in the hospital on day 120. Perhaps if the VA- NSQIP calculation was available preoperatively (as it already is for cardiac surgery) by entry of the age of a patient, his or her required laboratory study results, ASA score, and so on into a Web site or smartphone application, the surgeon, patients, and families of such patients might opt for palliative measures instead of surgery. However, the potential usefulness of the VA-NSQIP–calculated probability of death in preoperative informed consent must be studied prospectively before such a recommendation could be made.
Quantification of mortality after any given operation is used in a number of ways. With respect to quality improvement, mortality rates can be used as instruments of progress within an individual hospital or medical system. The VA-NSQIP is a notable success story in this regard, and there is little doubt that the quality of surgical care in the VA system has improved since the implementation of the VA-NSQIP in 1994.10 Similarly, mortality rates are often used to compare outcomes across the whole spectrum of hospitals in an effort to determine predictors of quality. In these settings, mortality at 30 days has proven an adequate data point. Although it is a somewhat arbitrary time point and might not reflect the true mortality after an operation, it serves as a good “yardstick” with which to compare various hospitals. Birkmeyer and colleagues25 address this issue explicitly in their analysis of the relationship between hospital volume and outcome for 2.5 million patients (14 different operations) drawn from the Nationwide Inpatient Sample: “While a large proportion of surgical deaths before discharge occurred more than 30 days after surgery, we decided that 30-day mortality alone would not adequately reflect true operative mortality . . . [However] associations between volume and outcome were largely unchanged when we repeated the analysis using 30-day mortality alone.”25(p1129) Therefore, the fact that mortality after colorectal surgery continues to increase after the 30-day mark in this small subset of patients whose data is contained in the VA-NSQIP database (assuming this is true of the wider VA-NSQIP population) does not likely detract from the effectiveness of the VA-NSQIP as a tool for quality improvement.
Mortality rates are at times used to trumpet the successes of individual hospitals26 or surgeons27 who cite low or even zero mortality after complex operations. At times, it is difficult to even find the definition of postoperative in these articles. Moreover, there is growing recognition that for a spectrum of operations, 30-day mortality significantly underestimates the true risk associated with the operation.12- 14 For example, Mullen and colleagues,13 who described the outcomes after hepatic resection among more than 1000 patients at 4 different institutions, found that mortality increased by 47% when the definition of postoperative was extended from 30 to 90 days. The authors therefore recommend that the standard definition of postoperative mortality after liver resection should include all deaths to a minimum of 90 days. Our results echo this conclusion, albeit for an entirely different group of operations. The 30-day mortality rate after colorectal surgery considerably underreports the true risk of dying after the operation. The 90-day morality rate proved a more accurate representation of the period within which patients died of postoperative complications.
Finally, mortality rates are certainly a critical component of the discussion about informed consent between the surgeon and patient that precedes any operation.2 There is evidence that surgeons likely already understate the mortality rates associated with many operations because of the publication bias in the literature.28 Mortality rates are a bit higher in general practice on a national level than is reported in small case series from centers of excellence. Our data suggest that the use of these oft-quoted 30-day mortality rates further understates the true risk of dying after surgery because the 30-day mark prematurely truncates the postoperative period. A significant portion of patients die after the 30-day mark of causes related to the operation. If, as surgeons obtaining consent before a routine elective colectomy, we quote a 1.4% risk of dying from the operation (the 30-day mortality for elective operations in this series), we would clearly be underreporting the risk of death, given that it increases to 4.1% at 90 days.
This series is a small cohort from a single institution, and the VA population is not entirely reflective of the broader population. Typically, VA patients are older and have more comorbidities, so they may be at higher risk of death at a later time after surgery. However, these data suggest that further investigation of mortality after 30 days among patients in the larger national VA-NSQIP database is warranted. In addition, when possible, surgical investigators in all settings should include both 30-day and 90-day mortality rates when they report outcomes.
In conclusion, in this series of patients who undergo colorectal surgery, 30-day mortality rates significantly underreport the true risk of death after surgery. Ninety-day mortality rates should be included as a standard outcome measure after CRS because they serve as a better estimation of risk when counseling patients.
Correspondence: Brendan C. Visser, MD, Palo Alto Veterans Health Care System, 3801 Miranda Ave, Palo Alto, CA 94304 (Brendan.Visser@va.gov).
Accepted for Publication: October 6, 2008.
Author Contributions:Study concept and design: Visser and Wren. Acquisition of data: Visser, Keegan, Martin, and Wren. Analysis and interpretation of data: Visser and Wren. Drafting of the manuscript: Visser, Martin, and Wren. Critical revision of the manuscript for important intellectual content: Visser, Keegan, and Wren. Statistical analysis: Visser. Administrative, technical, and material support: Visser, Keegan, Martin, and Wren. Study supervision: Wren.
Financial Disclosure: None reported.
Previous Presentations: Presented at the 77th Annual Meeting of the Pacific Coast Surgical Association; February 16, 2008; San Diego, California.
Additional Contributions: Alex McMillan, PhD, provided assistance with statistical analysis.